Data processing system with data transmit capability

A data processing system with data transmit capability comprising an operating system for supporting processes, such that the process are associated with one or more resources and the operating system being arranged to police the accessing by processes of resources so as to inhibit a process from accessing resources with which it is not associated. Part of this system is an interface for interfacing between each process and the operating system and a memory for storing state information for at least one process. The interface may be arranged to analyze instructions from the processes to the operating system, and upon detecting an instruction to re-initialize a process cause state information corresponding to that pre-existing state information to be stored in the memory as state information for the re-initialized process and to be associated with the resource.

Skip to: Description  ·  Claims  ·  References Cited  · Patent History  ·  Patent History
Description
1. PRIOR APPLICATION DATA

This application is a divisional application of and claims priority to U.S. patent application Ser. No. 11/900,377 entitled User-Level Re-Initialization Instruction Interception filed Sep. 10, 2007, which is a continuation of and claims priority to PCT Application No. PCT/GB2006/000852 filed on Mar. 10, 2006, which claims priority to Great Britain application number 0505300.4 filed Mar. 15, 2005.

2. FIELD OF THE INVENTION

This disclosure relates to data processing systems.

3. RELATED ART

In the field of computer systems it is generally necessary to determine an appropriate path by which to transmit instructions between elements of a computer. Typically the path is defined by the operating system running on the computer, but it is known that other processes can intervene in some situations to alter the path that instructions take. For example an application which wishes to use a hot-pluggable input-output device will take an instruction path which is determined by a hot-plug controller device according to the particular device which is resident in the system at that time.

For example, the application can invoke a system call (syscall) for transmission of data through the socket and then via the operating system to the network. Syscalls can be thought of as functions taking a series of arguments which cause execution of the CPU to switch to a privileged level and start executing the operating system. A given syscall will be composed of a specific list of arguments, and the combination of arguments will vary depending on the type of syscall.

Syscalls made by applications in a computer system can indicate a file descriptor (sometimes called a Handle), which is usually an integer number that identifies an open file within a process. A file descriptor is obtained each time a file is opened or a socket or other resource is created. File descriptors can be re-used within a computer system, but at any given time a descriptor uniquely identifies an open file or other resource. Thus, when a resource (such as a file) is closed down, the descriptor will be destroyed, and when another resource is subsequently opened the descriptor can be re-used to identify the new resource. Any operations which for example read from, write to or close the resource take the corresponding file descriptor as an input parameter. Existing data processing system suffer from various drawbacks in this regard.

In addition, in order to transmit data between data processors in a network such as an Ethernet network, data is formed into packets. Each packet includes a header specifying the destination of the data in the packet. In an Ethernet network the destination is typically specified by means of an Ethernet address, an Internet Protocol (IP) address and a Transmission Control Protocol (TCP) address.

In known network systems it is common for network routing rules to be stored in tables or other data structures such that when a transmission of data is requested the tables can be accessed to determine the appropriate addresses to which to send the data, and the contents of the packet headers can thus be created. Such tables are generally stored by the operating system of the terminal device that is to send the data, for example a personal computer or server.

Layers of the stack include an application and a socket provided by a socket library. The socket library is an application program interface (API) for building software applications. The socket library can carry out various functions, including creating descriptors and storing information. Additionally, there is an operating system (OS) comprising a TCP kernel and a proprietary TCP user-level stack.

In order to perform routing the user-level stack must use a routing table. One option would be for the user-level stack to maintain its own routing tables independently of the OS. However, this would require the user-level stack (in addition to the OS) to have access to all the communications necessary for establishing a routing table. There would then be ambiguity as to where those communications should be directed. Alternatively, the user-level stack may be able to access the routing tables stored by the OS. Since the user-level stack may have to access the tables very frequently during operation, accessing the routing tables stored by the OS is likely to create a significant workload for the system and so it can be expected to be inefficient for an application to be required to access tables in the OS each time it wishes to transmit data across the network. This is a drawback to the prior art.

It is further known that in computer systems, regions of memory are commonly shared between more than one application. Applications which are running are known as processes, and more than one process in a computer may require access to the shared memory at the same time. However, the regions of memory will typically be controlled by means of an operating system which will provide support to enable only one application at a time to access the shared memory space, as discussed in more detail below.

Multiple threads can exist within a single application process and can execute concurrently with access to all the memory of the application context. Thus, there may be multiple threads within each application wishing to access the shared memory. If more than one process or thread were permitted concurrent access to the memory then the application would be likely to crash since the same region of memory cannot be modified simultaneously by more than one set of instructions. Therefore, it is known to provide a lock associated with the memory. The lock can be changed between an unlocked state when no application is accessing the region of memory and a locked state when the memory is being accessed. Thus, when one thread (L) has access to the shared memory, the lock associated with the memory will enter a locked state, indicating that the memory cannot currently be accessed by other threads. When another thread (T) makes an attempt to access the memory while the thread L has access, the thread T will not be permitted access and will need to wait until the memory becomes available.

Spin-locks are commonly used by processes attempting to access shared memory. When a process makes an attempt to access the memory the process will either obtain the lock or it will not. If it fails, a decision must be made about how to proceed. If the process cannot proceed with any other operations until the lock is obtained then it will block and repeatedly attempt to access the memory until the lock is finally obtained. This can obviously be inefficient. An alternative is for the process to request a callback, so that when the lock is released the process is woken and can then re-try the lock. Although this can eliminate the requirement for a process to continually try the lock, it can still be inefficient because the process may not be able to carry out other operations while waiting for the memory access. In other words, it may have to block while waiting for a wake-up from the operating system.

In known systems, attempts made by threads to enter the memory space while it is being used can cause an entry to be added to a queue so as to indicate that the threads are awaiting access to the memory. If the memory is busy then, when it subsequently becomes available, a “wake-up” call can be sent to any waiting threads or applications. The waiting threads or applications are thus alerted to the availability of the memory and can then each make a further attempt to access the memory. Typically, the first thread or application to attempt to access the memory will be given access, although other scheduling algorithms are employed where appropriate. When an application successfully engages the memory, the lock will enter a locked state and access to the memory will be prevented to other applications.

It is also known for an application, on releasing a lock, to send a system call (syscall) to a driver within the operating system to initiate the sending of a wake-up call to waiting applications. The driver on receiving a syscall instructing a wake-up call, would access the queue to determine which applications are waiting, and send wake-up calls to the appropriate applications.

This arrangement has the following disadvantages. First, it can be inefficient for applications to have to make repeated attempts to access the memory. Second, some applications will have a higher priority than others and it can be very inefficient for a high priority application to be made to wait and make several access attempts before being permitted access to the memory. For example, an application may be blocked until it can access the memory, and it can therefore be important for that application to be treated as a high priority. Also the priority of all the waiting threads cannot be easily expressed in terms of the Operating System level inputs to the scheduling algorithm and is only known (or expressed) by the particular application and driver code.

Various embodiments are disclosed herein for overcoming the drawbacks in the prior art and for providing additional advantages and benefits for data processing systems and the methods associated therewith.

SUMMARY

To overcome the drawbacks of the prior art and provide additional benefits and features, a data processing system is disclosed. In one embodiment the system comprises an operating system for supporting processes such that each process is associated with one or more resources. The operating system is arranged to police the accessing by processes of resources so as to inhibit a process from accessing resources with which it is not associated, and being responsive to instructions of a certain type to re-initialize a process. Also part of this embodiment is an interface for interfacing between each process and the operating system. A memory is provided for storing state information for at least one process such that the state information is associated with a resource. In addition, the interface is arranged to analyze instructions from the processes to the operating system, upon detecting an instruction to re-initialize a process so as to be associated with a resource that is associated with pre-existing state information. This causes state information corresponding to that pre-existing state information to be stored in the memory as state information for the re-initialized process and to be associated with the resource.

Also disclosed herein is a data processing system comprising an operating system that stores a first network routing table that comprises one or more entries each indicating at least part of a route over a network to a respective destination. This system also comprises a transmission support function arranged to maintain a copy of the first network routing table and a network data transmission function arranged to transmit data in accordance with the copy network routing table and without accessing the first network routing table.

In one embodiment, a data processing system is arranged to control access by a plurality of processes to a region of shared memory. In such an embodiment, the data processing system is configured to prevent more than one process from concurrently accessing the region of shared memory and establish a data structure for holding items of a first type. These each comprise an indication of another of the plurality of processes, which are attempting to access the region of shared memory, and a definition of an operation on the region of shared memory. In addition, on a process finishing accessing the region of shared memory, action an item in the data structure by performing the defined operation by other than the process indicated in that item.

Also disclosed herein is a data processing system comprising an operating system for supporting processes such that the operating system is arranged to output data by addressing the data to a descriptor corresponding to the intended destination of the data. Also part of this embodiment is an interface for interfacing between each process and the operating system. A memory is provided including one or more address spaces and there is an address space corresponding to each of the processes for use by the respective process. Also, a data communication stack is provided or created for one of the processes such that the stack is located in the address space corresponding to the respective process and having a descriptor allocated to it. In this embodiment, the interface is arranged to analyze requests to alter the allocation of a descriptor sent by the processes towards the operating system to identify whether such a request relates to altering the allocation of the descriptor allocated to the stack, and if it does, allocate a different descriptor to the stack and subsequently permit the request to be processed.

Also disclosed herein is a method of compiling a data structure in a computer system such that the computer system is arranged to perform protocol processing and transmit data via an interface based on routing instructions. This method may be performed by a socket library, such as a socket library containing a TCP/IP stack, within the computer system. In addition, a program may be used to cause or program the computer system to compile the data structure in accordance with the method. A computer system programmed in this manner, and a data carrier carrying the program is also disclosed.

In one embodiment, the method comprises receiving a first instruction and determining the route indicated by the first instruction. The first instruction also identifies a descriptor which uniquely identifies an open file, socket, or other resource. Thus, the first instruction may be an instruction which provides a route to transmit data between elements of a computer such as a system call. For example, the first instruction could be a system call for transmission of data through a socket.

In this embodiment, an interface within the determined route is identified and an attempt is made to determine ownership of the interface. The interface may be a port on a network interface card. Ownership typically refers to the allocation of a resource, such as the interface, to an entity, such as a process, and may imply that access to that resource is limited to that entity. In some embodiments, ownership of the related descriptor may determine ownership of the interface. In addition, the ownership determination may be based on system-wide rules defining routes for transmitting data.

An entry in the data structure is created to store the result of the attempt to determine ownership along with the descriptor's identifier. The data structure is comprised of these entries and additionally provides routing information related to the descriptors within these entries to socket libraries or other libraries. In one or more embodiments, the state of the descriptor may indicate ownership by the operating system's kernel, ownership by another entity, or unknown ownership. The entries are used to efficiently identify ownership of a particular descriptor for subsequent instructions. For example, once an entry for a descriptor has been made, a subsequent instruction may be quickly routed to the kernel or another entity simply by accessing the descriptor's entry and reading its state.

The entries in the data structure may become obsolete for various reasons including but not limited to closing the descriptor, policy changes, or route updates caused by changes to network topology. In these cases, an indication that the descriptor's ownership may have changed is received, and the descriptor's entry in the data structure is changed to indicate its ownership is unknown. A subsequently received instruction identifying the descriptor will then cause the descriptor's entry to be updated with the determined ownership of an identified interface. As above, the interface is identified by determining the route specified by the instruction and identifying an interface within that route.

A computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route may route data by utilizing the above described data structure. When an instruction indicating a descriptor is received, the data structure is accessed to determine the indicated descriptor's state (i.e. ownership). The instruction is then directly transmitted according to a corresponding route without the high overhead of first being sent to the operating system. In one or more embodiments, routing may be performed by a socket library and the data structure may be stored within the socket library if desired.

Other systems, methods, features and advantages of the invention will be or will become apparent to one with skill in the art upon examination of the following figures and detailed description. It is intended that all such additional systems, methods, features and advantages be included within this description, be within the scope of the invention, and be protected by the accompanying claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention. In the figures, like reference numerals designate corresponding parts throughout the different views.

FIG. 1 shows a prior art computer system.

FIG. 2 represents a series of operations in a computer system.

FIG. 3 shows a computer system in accordance with embodiments of the present invention.

FIG. 4 shows a descriptor table.

FIG. 5 shows an arrangement of tables in accordance with an embodiment of the present invention.

 DETAILED DESCRIPTION

The embodiments of the invention described herein may include any one or more of the features described in relation to other inventions. As such various different embodiments may be configured with any element, feature, or step, disclosed herein, either alone or in combination.

This invention relates to the transmission of data in computer systems. In the field of computer systems it is generally necessary to determine an appropriate path by which to transmit instructions between elements of a computer. Typically the path is defined by the operating system running on the computer, but it is known that other processes can intervene in some situations to alter the path that instructions take. For example an application which wishes to use a hot-pluggable input-output device will take an instruction path which is determined by a hot-plug controller device according to the particular device which is resident in the system at that time.

FIG. 1 represents equipment capable of implementing a prior art protocol stack, such as a transmission control protocol (TCP) stack in a computer connected to a network. The equipment includes an application 1, a socket 2 and an operating system 3 incorporating a kernel 4. The socket connects the application to remote entities by means of a network protocol, in this example TCP/IP. The application can send and receive TCP/IP messages by opening a socket and reading and writing data to and from the socket, and the operating system causes the messages to be transported across the network. For example, the application can invoke a system call (syscall) for transmission of data through the socket and then via the operating system to the network.

Syscalls made by applications in a computer system can indicate a file descriptor (sometimes called a Handle), which is usually an integer number that identifies an open file within a process. A file descriptor is obtained each time a file is opened or a socket or other resource is created. File descriptors can be re-used within a computer system, but at any given time a descriptor uniquely identifies an open file or other resource. Thus, when a resource (such as a file) is closed down, the descriptor will be destroyed, and when another resource is subsequently opened the descriptor can be re-used to identify the new resource. Any operations which for example read from, write to or close the resource take the corresponding file descriptor as an input parameter.

When a network related application program interface (API) call is made through the socket library this causes a system call to be made, which creates (or opens) a new file descriptor. For example the accepts system call takes as an input a pre-existing file descriptor which has been configured to await new connection requests, and returns as an output a newly created file descriptor which is bound to the connection state corresponding to a newly made connection. The system call when invoked causes the operating system to execute algorithms which are specific to the file descriptor. Typically there exists within the operating system a descriptor table which contains a list of file descriptors and, for each descriptor, pointers to a set of functions that can be carried out for that descriptor. Typically, the table is indexed by descriptor number and includes pointers to calls, state data, memory mapping capabilities and ownership bits for each descriptor. The operating system selects a suitable available descriptor for a requesting process and temporarily assigns it for use to that process.

Certain management functions of a computing device are conventionally managed entirely by the operating system. These functions typically include basic control of hardware (e.g. networking hardware) attached to the device. When these functions are performed by the operating system the state of the computing device's interface with the hardware is managed by and is directly accessible to the operating system. An alternative architecture is a user-level architecture, as described in WO 2004/079981 and the applicant's copending PCT application claiming priority from GB 0408876.1. In a user-level architecture at least some of the functions usually performed by the operating system are performed by code running at user level. In a user-level architecture at least some of the state of the function can be stored by the user-level code. This can cause difficulties when an application performs an operation that requires the operating system to interact with or have knowledge of that state.

In embodiments of the present invention syscalls passing through the socket can be analyzed to establish the file descriptor and any information identified in the syscall that indicates the path by which the syscall is to be directed, and decisions can thereby be made by the socket so that the syscall can be transmitted in a suitable way from the socket.

An example of a syscall is Dup2(a,b), which has the effect of duplicating the file or other resource represented by descriptor “a” and creating a new resource represented by descriptor “b” and having the same properties. One example of when such a call might be useful is when a descriptor that has a system-wide significance (for example the descriptor that maps on to error output—commonly descriptor #2) is to be redirected on to some other file or device.

Other examples of syscalls are fork( ) and exec( ). A fork( ) call typically creates a new process (child) from the old one (parent) which initially shares all state including memory mappings and file-descriptors. After a successful fork( ), two copies of the original code will be running. An exec( ) call can then be requested for the child process. This will replace the current process image with a new process image, but details of the child process can be preserved. For example, specific file descriptors can be preserved in the child and (often) closed by the parent; thus handing over control of a file descriptor from a parent to a new child process.

According to a first aspect of the present invention there is provided a method of compiling a data structure in a computer system, the computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route, the method comprising: receiving a first instruction including an indication of a descriptor; determining a route indicated by the first instruction; identifying an interface within the determined route; attempting to determine the ownership of the interface; and causing an entry to be made in the data structure, the entry including an identifier of the descriptor and a state of the descriptor, wherein the state represents the result of the attempting step.

The method is preferably performed by a socket library in the computer system. The socket library preferably contains a TCP/IP stack.

The state may be a state from the list comprising A, B and C, where: A indicates ownership by a kernel of an operating system in the computer system; B indicates ownership by another owner; and C indicates unknown ownership. The first instruction could suitably be a system call. The descriptor is preferably a file descriptor. The determined ownership of the interface could conveniently be taken to be the ownership of the descriptor. The step of attempting to determine the ownership of the interface preferably includes accessing system-wide rules defining routes for transmitting data.

The method could further comprise: receiving a second instruction including an indication of the descriptor; accessing the data structure to determine the state of the descriptor; and routing the second instruction based on the determined state of the descriptor.

The method could additionally comprise: receiving an indication that the ownership of the descriptor may have changed; and changing the entry in the data structure to indicate the state of the descriptor as C.

The method could further comprise updating the state of the descriptor in the data structure by: receiving a third instruction including an indication of the descriptor; determining a route specified by the third instruction; identifying an interface within the determined route; attempting to determine the ownership of the interface; and based on the determined ownership of the interface, updating the said entry in the data structure to indicate the current ownership of the descriptor.

According to a second aspect of the present invention there is provided a data structure accessible by a socket library in a computer system, the data structure comprising one or more entries each including an identifier of a descriptor and a state of the descriptor, wherein the state represents the ownership of the descriptor, and wherein accessing the data structure can provide routing information relating to the descriptor.

According to a third aspect of the present invention there is provided a method of routing data in a computer system, the computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route, the method comprising: receiving a first instruction including an indication of a first descriptor; accessing a data structure storing, for each of one or more descriptors, an identifier of the descriptor and a state of the descriptor, wherein the state represents the ownership of the descriptor; determining the state of the first descriptor; in accordance with the determined state, determining a route; and transmitting the first instruction in accordance with the determined route. The method is preferably performed by a socket library in the computer system. The data structure is preferably stored at the socket library.

According to a fourth aspect of the present invention there is provided a program for programming a computer system to compile a data structure, the computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route, the program being arranged to cause the computer system to compile the data structure by a method comprising: receiving a first instruction including an indication of a descriptor; determining a route indicated by the first instruction; identifying an interface within the determined route; attempting to determine the ownership of the interface; and causing an entry to be made in the data structure, the entry including an identifier of the descriptor and a state of the descriptor, wherein the state represents the result of the attempting step.

According to a fifth aspect of the present invention there is provided a computer system programmed with a program as defined above. According to a sixth aspect of the present invention there is provided a data carrier carrying a program as defined above.

FIG. 3 shows components implementing a TCP stack for use in accordance with embodiments of the present invention. Layers of the stack include an application 1 and a socket 2 provided by a socket library. The socket library is an application program interface (API) for building software applications. The socket library can carry out various functions, including creating descriptors and storing information. Additionally, there is an operating system 3 comprising a TCP kernel 4, and a proprietary TCP user-level stack 5. It will be understood by the skilled person that although TCP is referred to by way of example, other protocols could also be used in accordance with embodiments of the invention. For example, User Datagram Protocol (UDP), Internet Control Message Protocol (ICMP) or Real-Time Transport Protocol (RTP) could be used. Non-Ethernet protocols could be used. The user-level stack is connected to hardware 6 in FIG. 3. The hardware could be a network interface card (NIC). The system shown in FIG. 3 is capable of performing protocol processing. This involves the processing of data packets and of the protocols according to which the data packets are transmitted within computer networks.

In this arrangement there can be one user-level TCP stack 5 for each application that requires one. This can provide better performance than if a stack is shared between applications. Each stack is located in the same address space as the application that it serves.

The socket library maintains a table 40, shown in FIG. 4, incorporating identifiers of file descriptors and their ownership. In general, ownership applies to control of access to elements within a computer system. For example, a network interface 6 (typically a port on a network interface card) could link a data processing system to a series of other computers, and the data processing system could be similarly linked by a further network interface to another series of computers. If it is desired to send a packet of data from the data processing system to a specific one of the computers, the correct network interface must be selected in order to successfully transmit the packet to the correct computer. In this case, the term “ownership” refers to the identity of the interfaces. Thus “ownership” typically refers to the allocation of a resource to an entity such as a process, which may imply that access to that resource is limited to that entity.

In the present example illustrated by FIG. 3, the computer system has a kernel (K) 4 and a proprietary user-level stack 5 which will be referred to (by way of example only) as a Level 5 (or L5) stack. The L5 stack is associated with its own library which is interposed in the system. The ownership of file descriptors in use in the system is defined according to which network interface the file descriptor is associated with. In this case, the ownership can have three values in: L5; K; or unknown/indeterminate. These values could be indicated explicitly or by way of binary flags. The descriptor table maintained by the socket library indicates whether each descriptor is owned by L5, owned by K, or of unknown ownership. Exemplary entries in the table 40 are shown in FIG. 4. The descriptor numbers are listed in the left column and an indication of the ownership of each descriptor is shown in the right column. Thus, the table shows that descriptor number 0 has been determined to be owned by L5, descriptor numbers 3 and 6 are of unknown ownership and descriptor number 4 is owned by the kernel. The table 40 is preferably stored securely such that users cannot access it directly and corrupt it by changing pointers in the table.

In FIG. 2, a series of operations is illustrated. An application 1 invokes a socket( ) syscall 10 through the socket API 2, requesting that a new file descriptor be created. This could alternatively be achieved for example by using an open( ) call. In the present example, the application could be a webserver which creates a new descriptor which accepts new connections and then forks( ) a new process for each concurrent user.

In the example, the server is initializing, and so a socket( ) call is made by the application.

At step 11 the socket library, which may be a standalone library or alternatively its functionality could be incorporated in a single system library such as libc invokes a syscall trap which causes execution to switch to the operating system. The operating system determines the syscall source and executes internal socket creation code within its network subsystem. This code will request a new file descriptor. The operating system checks its descriptor table and selects a descriptor D suitable for this application and then assigns it to the new file associated with the new user. An identifier of the selected descriptor D is then sent in step 12 by the operating system 3 to the socket 2.

In prior art systems, the passing of the syscall from the application to the socket library, and the invoking of the resulting system call at the socket is generally unhindered. In contrast, in the present system, the call is intercepted before it reaches the socket library. The intercepted message 12 is checked by the L5 socket library, and an identifier of the descriptor D is entered in the table 40, with an indication of indeterminate ownership. An extract of the table 40 is shown below step 12 in FIG. 2.

An identifier of the descriptor D in the message 12 is then transmitted to the requesting application 1 in step 13. The application may then include this descriptor in further syscalls which it transmits, enabling the relevant file to be identified.

Syscalls can be thought of as functions taking a series of arguments which cause execution of the CPU to switch to a privileged level and start executing the operating system. Here the syscalls are denoted 1 to N. A given syscall will be composed of a specific list of arguments, and the combination of arguments will vary depending on the type of syscall.

In this embodiment, a network route table is maintained by the operating system 3 storing arguments and their associated routes including interfaces within the routes. The network route table typically stores network subnet address masks and their associated network interfaces. A given network address is matched against the table in such a way that the most specifically defined route can be chosen.

This table contains all external network routing rules for the computer system. The table is also associated with an API which enables entries in the table to be added or deleted. Because the route table is held in the OS a user-level stack would waste time making system calls to determine the route of packets being transmitted. To avoid this a local copy of the route table (and other control plane tables such as the list of network interfaces and the address resolution protocol (ARP) table) is maintained in the context of the user-level application. In the system of this example, the L5 stack is registered with the operating system to receive updates when the route table changes. The table is thereby copied into the L5 stack and if an application's route changes, the relevant information will be relayed to the L5 stack.

The L5 user-level stack provides a “look up route” function which, on receiving a request, can return details of whether or not a specified route goes through an interface associated with the L5 stack (in this example an L5 interface). This function will check a generation count to determine whether its cached route table state is still valid. If so it can use cached state, otherwise it needs to either make a system call or consult shared memory pages onto valid route table entries. Thus, a change can be indicated in a two-step process. First incrementing a “generation counter” means that the socket library knows quickly that something has changed. Then, once off the fast path, the socket library can determine the new state.

In step 14, another syscall is sent from the application 1. In this example it is a connect( ) syscall, specifying an address for connection which could be in another machine. The socket library 2 intercepts the message 14, determines the type of syscall and looks it up in a further table. If it is a type from which it will not be possible to determine the ownership of a descriptor, no further processing of the message is performed to establish the ownership. An entry already exists in the table 40 for the descriptor D, and the ownership indicated in the table will remain unchanged, as indeterminate. Information and the descriptor D in the message 14 will then be checked by the socket 2 to establish whether the message should be passed to the kernel or the L5 stack, and it will then be transmitted within the system to the appropriate interface.

On the other hand, if the syscall 14 is determined by the socket library to be of a type that could identify the ownership of the descriptor, the syscall will be further analyzed. In the present example, the message is a connect request and so it will be further analyzed. This analysis includes identifying the descriptor D and any arguments included in the message. The arguments can then be analyzed by the socket library by means of a “look up route” request. The route table copied into the L5 stack will then be checked and an associated interface can thereby be identified. The ownership of the descriptor D is assumed to be the same as that of the interface, and the ownership data in the table 40 against descriptor D can then be updated. The socket library can thus identify whether the descriptor should be passed to the operating system 3 (or, more specifically, the kernel 4) or to the proprietary user-level stack 5. In the present example, the syscall 14 is determined to be directed to a L5 interface, and the descriptor D is therefore taken to be a L5 descriptor. The table 40 is updated with a positive indication of L5 ownership, as shown below step 14 in FIG. 2, and the syscall will be passed to the L5 stack which will perform the required operation within the application context.

When a subsequent message identifying descriptor D passes through the socket 2, the socket library can simply determine the descriptor from the message and look up that descriptor in the locally stored table 40 to determine its presumed ownership. Messages incorporating a descriptor owned by L5 will be intended to be passed directly down from the socket 2 to the L5 user-level stack 5 shown in FIG. 3. Because the ownership of the descriptor D is determined from the table 40 as being L5, messages intercepted by the socket library incorporating an identifier of the descriptor D will be passed directly down to the L5 stack 5 without first being sent to the operating system 3. Therefore, by means of embodiments of the present invention, only a simple analysis of a message passing through the socket library is required in order to be able to establish the appropriate part of the stack to which to pass the message. The high overhead in processing instructions to determine a path is thereby avoided. Preferably, if the ownership of a descriptor is indeterminate, any message incorporating that descriptor will be sent to the operating system by default.

When a file is to be closed, a destructor syscall (usually close( )) is sent from the application 1 and intercepted by the socket library 2. The socket library can then identify that the descriptor is to be destroyed and remove the entry for that descriptor from the table 40. Then, when the descriptor is subsequently re-used by the operating system and assigned to a new process, the socket library can intercept the message from the operating system identifying the newly-assigned descriptor, as described above in relation to step 12 of FIG. 2. Alternatively, the entry could remain in the table and the associated indication of its ownership could be modified to show that its ownership is unknown. Alternatively, the default setting for a descriptor's ownership could be “operating system”.

The information stored in the table 40 may become obsolete when routing instructions within the computer system change. This could happen as a result of policy changes, for example when an application no longer runs with Level 5, or, more commonly, as a result of load balancing decisions within the computer system or other route updates caused by network topology changes. The information in the table 40 should be checked when such a change occurs. A convenient way of arranging the checking procedure is to reset a descriptor's ownership to indeterminate (or K) when such a change occurs so that until the correct new ownership information can be obtained a full processing operation will be carried out for the routing of messages to the operating system or the L5 stack via the socket 2. More generally a single generation counter is associated with each user-level socket state. Incrementing this counter will cause the L5 user stack to leave its fast path processing and determine the state change

In a typical system as shown in FIG. 3, each application 1 has its own user-level TCP stack 5 by which it can communicate with its hardware 6. This enables shorter processing times because the application need not share the stack with other applications. As stated above, the stack 5 is in the same address space as its corresponding application. Additionally, each stack 5 is associated with a dedicated driver (not shown) in the kernel 4 of the operating system. In this arrangement, when the application 1 attempts to open a descriptor that is being used by the stack 5, for example a descriptor for accessing the TCP driver in the kernel 4, the operating system 3 cannot identify that there is a conflict because the stack 5 and the application 1 are in the same address space. The conflict arises due to the fact that in order for the L5 stack to operate it requires operating system descriptors, and these descriptors must be allocated from the same address space as the descriptors used by the application. However, the application has no a prior knowledge of the existence of the L5 stack.

A specific example will now be described. The operating system 3 allocates descriptor number 42 for use by the L5 TCP stack 5. The application 1 then sends a Dup2(X,42) call, requesting that descriptor 42 becomes a copy of descriptor X. If this instruction were executed by the operating system it would cause a conflict because descriptor 42 is already in use to identify the stack. Normally such a conflict would be avoided by the operating system preventing one process from having access to a descriptor that is already in use by another process. However, in this case the application and the user-level stack 5 occupy the same application space as far as the operating system is concerned, and so the operating system could not normally allow this Dup2( ) call to proceed, unless the application were to first close the existing resource having descriptor 42. To avoid such a conflict, the socket library 2 intercepts Dup2( ) calls and identifies whether they request a descriptor assigned to the stack to be redefined. The socket library checks in the table 40, which additionally includes a flag indicating whether each descriptor is a private L5 user-level stack descriptor, and if it determines that there will be a clash, a series of operations is carried out by the socket library.

Thus, the socket library intercepts the syscall Dup2(X,42) from the application 1, which is an attempt to transform descriptor X into descriptor 42. The socket library checks the table 40 for a flag indicating that 42 is private to the L5 stack. It determines that it is, and blocks the application's thread, as would happen for any syscall. The library obtains a lock on descriptor 42 to prevent other actions being carried out in relation to it. The library then sends a Dup2(42,Y) call to the operating system, requesting that the current definition of descriptor 42 is duplicated at some unused descriptor Y, so that the stack can subsequently operate with descriptor Y.

If this call fails, for example due to an internal error within the stack 5, the socket library forges a message to the application 1 to indicate that the Dup2(X,42) call from the application failed. On the other hand, if the Dup2(42,Y) call succeeds, the application's thread is released, and the Dup2(X,42) call can be forwarded by the socket library to the operating system. When the socket library receives a response from the operating system indicating that the Dup2(42,Y) call was successful, it forwards a response to the application, which the application interprets as being confirmation that its Dup2(X,42) call succeeded. The application can then use descriptor 42 and the stack can use the new descriptor Y, and the potential conflict is thus prevented.

The sequence of operations described above can involve a high processing overhead, so to inhibit an application from requesting the use of a descriptor that is in use by the stack, it is preferred that such a descriptor is marked by the operating system as reserved. The operating system would then be unlikely to inform an application that such a descriptor is available and so it is unlikely that the application will request the use of such a descriptor.

The dup2( ) instruction is an example of an instruction to duplicate a descriptor that is used in the Unix and Linux operating systems. Other operating systems may be responsive to other instructions to perform functionally equivalent duplication of descriptors, and similar techniques can be applied to avoid conflicts in those operating systems too.

When a fork( ), exec( ) sequence is performed on an application, the child application can retain the file descriptors of the parent application in such a way that all details of the parent application except the file descriptors are destroyed. By virtue of fork/exec operations with respect to a single user-level transport library, multiple applications can eventually be mapped onto the same state in an operating system. This can lead to considerable contention for the lock, which can be inefficient as applications are generally required to spin or block while they await access to the state.

When a new application is created, it acquires a process ID (PID) which is unique within a host. According to one embodiment as shown in FIG. 5, an exec table 51 is maintained by the L5 operating system 5 in which each record has a PID, pointers to a system-wide file descriptor table 52 which contains pointers to state 53 in the operating system, and file descriptors that are used by that PID, for example having been preserved following an exec( ) call.

In one embodiment, the socket library 2 intercepts an exec( ) call and can read it to reconstruct state data for a new L5 process. When the library has intercepted the exec( ) call, an entry is made in the exec table 51 and the exec( ) call is processed. The new process then starts, and a new L5 library is set up. The init( ) of the new library involves checking via an access of the L5 file descriptor to the exec table 51 of the operating system 5, and the exec table 51 is recreated as a new table (not shown in FIG. 5) in the library. This table can record details of whether a file descriptor has been involved in an exec( ) operation. This table could be the same as the table 40, but preferably it is separate. For efficiency reasons this table can be reconstructed in a “lazy” manner as the new application attempts to access each file descriptor.

The system-wide table 52 maintained by the operating system 5 can keep track of file descriptors which have been preserved following a fork-exec operation. This system-wide table records which file descriptors are in use for which applications. Each file descriptor is assigned a reference number by the operating system 5 which is unique across the system. This can be necessary because two applications which do not communicate with each other can use identical file descriptor numbers without any conflicts occurring. However, it is convenient to be able to identify a particular file descriptor in use by a particular application with a unique number.

The system-wide table 52 also stores pointers for each file descriptor to state 53 within the system. The state associated with a file descriptor of a child process will typically be the same as the state associated with its parent process and the process of reconstruction requires that the newly exec-ed application makes system calls to the L5 driver in order to reattach to the shared state.

As stated above, it can occur that due to multiple fork-exec operations many processes come to be sharing the same state. This is inefficient since a process would have to await access to a lock on the state and this generally involves blocking. To reduce such contention, the operating system 5 preferentially assigns new memory space (a new network interface) to each forked application as it requests its first new file descriptor.

The applicant hereby discloses in isolation each individual feature described herein and any combination of two or more such features, to the extent that such features or combinations are capable of being carried out based on the present specification as a whole in the light of the common general knowledge of a person skilled in the art, irrespective of whether such features or combinations of features solve any problems disclosed herein, and without limitation to the scope of the claims. The applicant indicates that aspects of the present invention may consist of any such individual feature or combination of features. In view of the foregoing description it will be evident to a person skilled in the art that various modifications may be made within the scope of the invention.

While various embodiments of the invention have been described, it will be apparent to those of ordinary skill in the art that many more embodiments and implementations are possible that are within the scope of this invention. In addition, the various features, elements, and embodiments described herein may be claimed or combined in any combination or arrangement.

Claims

1. A method of compiling a data structure that is performed by a socket library in a computer system, the computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route, the method that is performed by the socket library comprising:

receiving a first instruction comprising a system call and an indication of a descriptor;
determining a route indicated by the first instruction;
identifying, in response to the determination, an interface within the determined route;
attempting to determine the ownership of the interface wherein determining ownership of the interface comprises determining an entity to which the interface is allocated; and
causing an entry to be made in the data structure stored in the socket library, the entry including an identifier of the descriptor and a state of the descriptor, wherein the state represents the result of the attempting step and indicates an entity to which the socket library should send instructions including an indication of the descriptor.

2. A method as claimed in claim 1, wherein the socket library contains a TCP/IP stack.

3. A method as claimed in claim 1, wherein the state is a state from the list comprising A, B and C, where:

A indicates ownership by a kernel of an operating system in the computer system;
B indicates ownership by another owner; and
C indicates unknown ownership.

4. A method as claimed in claim 3, further comprising:

receiving an indication that the ownership of the descriptor may have changed; and
changing the entry in the data structure to indicate the state of the descriptor as C.

5. A method as claimed in claim 4 further comprising updating the state of the descriptor in the data structure by:

receiving a third instruction including an indication of the descriptor;
determining a route specified by the third instruction;
identifying an interface within the determined route;
attempting to determine the ownership of the interface; and
based on the determined ownership of the interface, updating the entry in the data structure to indicate the current ownership of the descriptor.

6. A method as claimed in claim 1, wherein the descriptor is a file descriptor.

7. A method as claimed in claim 1, wherein the determined ownership of the interface is taken to be the ownership of the descriptor.

8. A method as claimed in claim 1, wherein the step of attempting to determine the ownership of the interface includes accessing system-wide rules defining routes for transmitting data.

9. A method as claimed in claim 1, further comprising:

receiving a second instruction including an indication of the descriptor;
accessing the data structure to determine the state of the descriptor; and
routing the second instruction based on the determined state of the descriptor.

10. A method of routing data in a computer system, the computer system being arranged to perform protocol processing and transmit data via an interface on the basis of instructions in accordance with a route, the method comprising:

receiving a first instruction comprising a system call and an indication of a first descriptor;
accessing a data structure storing, for each of one or more descriptors, an identifier of the descriptor and a state of the descriptor, wherein the state represents the ownership of the descriptor;
determining the state of the first descriptor including determining an entity to which the descriptor is allocated;
in accordance with the determined state, determining a route; and
transmitting the first instruction in accordance with the determined route, wherein the method is performed by a socket library in the computer system and the data structure is stored in the socket library and the state indicates an entity to which the socket library should send instructions including an indication of the descriptor.

11. A method as claimed in claim 10 wherein the state is a state from the list comprising A, B and C, where:

A indicates ownership by a kernel of an operating system in the computer system;
B indicates ownership by another owner; and
C indicates unknown ownership.

12. A method as claimed in claim 10, wherein the descriptor is a file descriptor.

13. A method as claimed in claim 10, wherein the determined ownership of the interface is taken to be the ownership of the descriptor.

14. A non-transitory computer readable storage medium having a program for programming a computer system to compile a data structure stored in a socket library thereon, the computer system comprising a processor configured to perform protocol processing and an interface configured to transmit data on the basis of instructions in accordance with a route, the program comprising instructions arranged to, when executed, cause the socket library in the computer system to compile the data structure by:

receiving a first instruction comprising a system call and an indication of a descriptor;
determining a route indicated by the first instruction;
identifying, in response to the determination, an interface within the determined route;
attempting to determine the ownership of the interface wherein determining ownership of the interface comprises determining an entity to which the interface is allocated; and
causing an entry to be made in the data structure that is stored in the socket library, the entry including an identifier of the descriptor and a state of the descriptor, wherein the state represents the result of the attempting step and indicates an entity to which the socket library should send instructions including an indication of the descriptor.

15. A computer system comprising:

an interface configured to transmit data on the basis of instructions in accordance with a route;
a processor in communication with a memory, the processor configured to execute machine readable code to perform protocol processing;
a memory storing the machine readable code, the machine readable code being non-transitory and configured to: receive a first instruction comprising a system call and an indication of a descriptor; determine the route, the route being indicated by the first instruction; identify, in response to the determination, an interface within the determined route; attempt to determine the ownership of the interface wherein determine ownership of the interface comprises determining an entity to which the interface is allocated; and cause an entry to be made in the data structure that is stored in the socket library, the entry including an identifier of the descriptor and a state of the descriptor, wherein the state represents the result of the attempt step and indicates an entity to which the socket library should send instructions including an indication of the descriptor.
Referenced Cited
U.S. Patent Documents
5272599 December 21, 1993 Koenen
5325532 June 28, 1994 Crosswy et al.
5535416 July 9, 1996 Feeney et al.
5550816 August 27, 1996 Hardwick et al.
5603020 February 11, 1997 Hashimoto et al.
5634124 May 27, 1997 Khoyi et al.
5742817 April 21, 1998 Pinkoski
5946189 August 31, 1999 Koenen et al.
6085217 July 4, 2000 Ault et al.
6098112 August 1, 2000 Ishijima et al.
6145088 November 7, 2000 Stevens
6160554 December 12, 2000 Krause
6304945 October 16, 2001 Koenen
6349035 February 19, 2002 Koenen
6415333 July 2, 2002 Vasell
6424621 July 23, 2002 Ramaswamy et al.
6438130 August 20, 2002 Kagan et al.
6484224 November 19, 2002 Robins et al.
6502203 December 31, 2002 Barron et al.
6530007 March 4, 2003 Olarig et al.
6560613 May 6, 2003 Gylfason et al.
6667918 December 23, 2003 Leader et al.
6687762 February 3, 2004 Van Gaasbeck et al.
6687849 February 3, 2004 Cherf
6718392 April 6, 2004 Krause
6728743 April 27, 2004 Shachar
6732211 May 4, 2004 Goyal et al.
6735642 May 11, 2004 Kagan et al.
6768996 July 27, 2004 Steffens et al.
6785888 August 31, 2004 McKenney et al.
6904534 June 7, 2005 Koenen
6950961 September 27, 2005 Krause et al.
6978331 December 20, 2005 Kagan et al.
7093158 August 15, 2006 Barron et al.
7099275 August 29, 2006 Sarkinen et al.
7103626 September 5, 2006 Recio et al.
7103744 September 5, 2006 Garcia et al.
7136397 November 14, 2006 Sharma
7143412 November 28, 2006 Koenen
7149227 December 12, 2006 Stoler et al.
7151744 December 19, 2006 Sarkinen et al.
7216225 May 8, 2007 Haviv et al.
7240350 July 3, 2007 Eberhard et al.
7245627 July 17, 2007 Goldenberg et al.
7254237 August 7, 2007 Jacobson et al.
7285996 October 23, 2007 Fiedler
7316017 January 1, 2008 Jacobson et al.
7346702 March 18, 2008 Haviv
7386619 June 10, 2008 Jacobson et al.
7403535 July 22, 2008 Modi et al.
7404190 July 22, 2008 Krause et al.
7502826 March 10, 2009 Barron et al.
7509355 March 24, 2009 Hanes et al.
7518164 April 14, 2009 Smelloy et al.
7551614 June 23, 2009 Teisberg et al.
7554993 June 30, 2009 Modi et al.
7573967 August 11, 2009 Fiedler
7580415 August 25, 2009 Hudson et al.
7580495 August 25, 2009 Fiedler
7617376 November 10, 2009 Chadalapaka et al.
7631106 December 8, 2009 Goldenberg et al.
7650386 January 19, 2010 McMahan et al.
7688853 March 30, 2010 Santiago et al.
7757232 July 13, 2010 Hilland et al.
7801027 September 21, 2010 Kagan et al.
7802071 September 21, 2010 Oved
7813460 October 12, 2010 Fiedler
7827442 November 2, 2010 Sharma et al.
7835375 November 16, 2010 Sarkinen et al.
7848322 December 7, 2010 Oved
7856488 December 21, 2010 Cripe et al.
7864787 January 4, 2011 Oved
7904576 March 8, 2011 Krause et al.
7921178 April 5, 2011 Haviv
7929539 April 19, 2011 Kagan et al.
7934959 May 3, 2011 Rephaeli et al.
7978606 July 12, 2011 Buskirk et al.
8000336 August 16, 2011 Harel
20020059052 May 16, 2002 Bloch et al.
20020112139 August 15, 2002 Krause et al.
20020129293 September 12, 2002 Hutton et al.
20020140985 October 3, 2002 Hudson
20020152226 October 17, 2002 Burnett
20020156784 October 24, 2002 Hanes et al.
20030007165 January 9, 2003 Hudson
20030058459 March 27, 2003 Wu et al.
20030063299 April 3, 2003 Cowan et al.
20030065856 April 3, 2003 Kagan et al.
20030081060 May 1, 2003 Zeng et al.
20030172330 September 11, 2003 Barron et al.
20030182465 September 25, 2003 Moir et al.
20030191786 October 9, 2003 Matson et al.
20030202043 October 30, 2003 Zeng et al.
20030214677 November 20, 2003 Bhaskar et al.
20040071250 April 15, 2004 Bunton et al.
20040141642 July 22, 2004 Zeng et al.
20040190533 September 30, 2004 Modi et al.
20040190538 September 30, 2004 Bunton et al.
20040190557 September 30, 2004 Barron
20040193734 September 30, 2004 Barron et al.
20040193825 September 30, 2004 Garcia et al.
20040210754 October 21, 2004 Barron et al.
20040210906 October 21, 2004 Beresnevichiene et al.
20040252685 December 16, 2004 Kagan et al.
20050008223 January 13, 2005 Zeng et al.
20050018221 January 27, 2005 Zeng et al.
20050038918 February 17, 2005 Hilland et al.
20050038941 February 17, 2005 Chadalapaka et al.
20050039171 February 17, 2005 Avakian et al.
20050039172 February 17, 2005 Rees et al.
20050039187 February 17, 2005 Avakian et al.
20050066333 March 24, 2005 Krause et al.
20050172181 August 4, 2005 Huliehel
20050219278 October 6, 2005 Hudson
20050219314 October 6, 2005 Donovan et al.
20050231751 October 20, 2005 Wu et al.
20060026443 February 2, 2006 McMahan et al.
20060045098 March 2, 2006 Krause
20060126619 June 15, 2006 Teisberg et al.
20060165074 July 27, 2006 Modi et al.
20060193318 August 31, 2006 Narasimhan et al.
20060228637 October 12, 2006 Jackson et al.
20060248191 November 2, 2006 Hudson et al.
20070188351 August 16, 2007 Brown et al.
20070220183 September 20, 2007 Kagan et al.
20080024586 January 31, 2008 Barron
20080109526 May 8, 2008 Subramanian et al.
20080115216 May 15, 2008 Barron et al.
20080115217 May 15, 2008 Barron et al.
20080126509 May 29, 2008 Subramanian et al.
20080135774 June 12, 2008 Hugers
20080147828 June 19, 2008 Enstone et al.
20080148400 June 19, 2008 Barron et al.
20080177890 July 24, 2008 Krause et al.
20080244060 October 2, 2008 Cripe et al.
20080301406 December 4, 2008 Jacobson et al.
20080304519 December 11, 2008 Koenen et al.
20090165003 June 25, 2009 Jacobson et al.
20090201926 August 13, 2009 Kagan et al.
20090213856 August 27, 2009 Paatela et al.
20090268612 October 29, 2009 Felderman et al.
20090302923 December 10, 2009 Smeloy et al.
20100088437 April 8, 2010 Zahavi
20100138840 June 3, 2010 Kagan et al.
20100169880 July 1, 2010 Haviv et al.
20100188140 July 29, 2010 Smelloy
20100189206 July 29, 2010 Kagan
20100265849 October 21, 2010 Harel
20100274876 October 28, 2010 Kagan et al.
20110004457 January 6, 2011 Haviv et al.
20110010557 January 13, 2011 Kagan et al.
20110029669 February 3, 2011 Chuang et al.
20110029847 February 3, 2011 Goldenberg et al.
20110044344 February 24, 2011 Hudson et al.
20110058571 March 10, 2011 Bloch et al.
20110083064 April 7, 2011 Kagan et al.
20110096668 April 28, 2011 Bloch et al.
20110113083 May 12, 2011 Shahar
20110116512 May 19, 2011 Crupnicoff et al.
20110119673 May 19, 2011 Bloch et al.
20110173352 July 14, 2011 Sela et al.
Foreign Patent Documents
0 598 510 May 1994 EP
620521 March 1995 EP
2272214 January 2011 EP
0148972 July 2001 WO
0235838 May 2002 WO
2008127672 December 2008 WO
2009136933 November 2009 WO
2010020907 July 2010 WO
2010087826 August 2010 WO
2011043768 April 2011 WO
2011053305 May 2011 WO
2011053330 May 2011 WO
Other references
  • W. Richard Stevens, UNIX network programing, 1990, Prentice hall, ISBN 0139498761, p. 56.
  • Friedrich Seifert, SCI Socket—A Fast Socket Implmentation over SCI, citeseerx.ist.psu.edu, 2006.
  • Marc E. Fiuczynski, et al. “Spine Reports, Talks, Papers”, http://www.cs.princeton.edu/˜mef/research/spine/reports/sigops98/index.html, May 19, 2006, 10 pages.
  • Hermann Hellwagner, Josef Weidendorfer, “SCI Sockets Library”, http://www.itec.uni-klu.ac.at/, http://wwwbodes.in.tum.de/, 20 pages.
  • Kevin Fall, et al., “Improving Continuous-Media Playback Performance with In-Kernel Data Paths”, 0-8186-5530-5/94, © 1994 IEEE, pp. 100-109.
  • Maged M. Michael, et al., “Relative Performance of Preemption-Safe Locking and Non-Blocking Synchronization on Multiprogramm3ed Shared Memory Multiprocessors”, 1063-7133/97, © 1997 IEEE, pp. 267-273.
  • Maged M. Michael, “Hazard Pointers: Safe Memory Reclamation for Lock-Free Objects”, IEEE Transactions on Parallel and Distributed Systems, vol. 15, No. 6, Jun. 2004, pp. 491-504.
  • “HP-UX Route Administration Manager (ramD)” © 2003 Hewlett-Packard Development Company L.P., 11 pages.
  • Tom Zanussi, et al., “relayfs: An Efficient Unified Approach for Transmitting Data from Kernel to User Space”, Linux Symposium 2003, pp. 519-532.
  • Cramming more components onto integrated circuits, Electronics, vol. 38, No. 8, pp. 114-117, 1965, Gordon E. Moore, Apr. 19, 1965.
  • Programming Semantics for Multiprogrammed Computations, Communications of the ACM, vol. 9, No. 3, pp. 143-155, Jack B. Dennis and Earl C. Van Horn, Mar. 1966.
  • The Aloha System, ACM Computer Communication Review, vol. 4 No. 1, F.F. Kuo, Jan. 1974.
  • A Protocol for Packet Network Intercommunication, IEEE Transactions on Communications, vol. COM-22, No. 5, pp. 637-648, Vinton Cerf, Robert Kahn, May 1974.
  • Proposal for an International End-to-End Protocol, ACM Computer Communication Review, vol. 6 No. 1, p. 1-18, V. Cerf, et al., Jan. 1976.
  • Ethernet: distributed packet switching for local computer networks, Communications of the ACM, vol. 19, Issue 7, pp. 395-404, Robert M. Metcalfe and David R. Boggs, Jul. 1976.
  • Interrupt Driven Programming, Communications of the ACM, vol. 14, No. 6, p. 417-418, Marvin Zelkowits, Jun. 1971.
  • Synchronizing Processors with Memory-Content-Generated Interrupts, Communications of the ACM, vol. 16, No. 6, p. 350-351, J. Carver Hill, Jun. 1973.
  • Virtual cut-through: A new computer communication switching technique, Computer Networks, vol. 3, No. 4, pp. 267-286, P. Kermani and L. Kleinrock, Sep. 1979.
  • An Overview of the New Routing Algorithm for the ARPANET, Proceedings of the 6th Data Communications Symposium, p. 63, John M. McQuillan, et al., Nov. 1979.
  • Grapevine: An Exercise in Distributed Computing, Communications of the ACM, vol. 25, Issue 4, pp. 260-274, Andrew D. Birrell, et al., Apr. 1982.
  • The Architecture of the Universe Network, ACM Computer Communication Review, vol. 14, No. 2, pp. 2-9, Ian M. Leslie, et al., Jun. 1984.
  • Congestion Control in IP/TCP Internetworks, ACM Computer Communication Review, vol. 14, No. 4, p. 11-17, John Nagle, Oct. 1984.
  • Development of a TCP/IP for the IBM/370, ACM Computer Communication Review, vol. 15, No. 4, Robert M. Brandriff, et al., Sep. 1985.
  • Supercomputers on the Internet: A Case Study, ACM Computer Communication Review, vol. 17, No. 5, C. Kline, Aug. 1987.
  • Fragmentation Considered Harmful, ACM Computer Communication Review, vol. 17, No. 5, pp. 390-401, Christopher A. Kent, Jeffrey C. Mogul, Oct. 1987.
  • An Analysis of Memnet: An Experiment in High-Speed Shared-Memory Local Networking, ACM Computer Communication Review, vol. 18, No. 4, p. 165-174, Gary S. Delp, et al., Aug. 1988.
  • Measured Capacity of an Ethernet: Myths and Reality, ACM Computer Communication Review, vol. 18, No. 4, p. 222-234, David R. Boggs, et al., Aug. 1988.
  • The VMP Network Adapter Board (NAB): High-Performance Network Communication for Multiprocessors, ACM Computer Communication Review, vol. 18, No. 4, p. 175-187, H. Kanakia and D. Cheriton, Aug. 1988.
  • Congestion Avoidance and Control, ACM Computer Communication Review, vol. 18, No. 4, p. 314-329, V. Jacobson, Aug. 1988.
  • The Design Philosophy of the DARPA Internet Protocols, ACM Computer Communication Review, vol. 18, No. 4, pp. 106-114, David D. Clark, Augus 1988.
  • Development of the Domain Name System, ACM Computer Communication Review, vol. 18, No. 4, pp. 123-133, Paul V. Mockapetris and Kevin J. Dunlap, Augus 1988.
  • Performance Comparison of the Cray-2 and Cray X-MP/416 Supercomputers, Proceedings of the 1988 ACM/IEEE conference on Supercomputing, p. 288-295, Orlando, Florida, Margaret L. Simmons and Harvey J. Wasserman, Nov. 12, 1988.
  • Implementing TCP/IP on a Cray computer, ACM Computer Communication Review, vol. 19, No. 2, p. 11-15, David A. Borman, Apr. 1989.
  • Computing the Internet Checksum, ACM Computer Communication Review, vol. 19, No. 2, p. 86-94, R. Braden, et al., Apr. 1989.
  • An Analysis of TCP Processing Overhead, IEEE Communications Magazine, vol. 27, No. 6, pp. 23-29, David D. Clark, et al., Jun. 1989.
  • Sirpent: A High-Performance Internetworking Approach, ACM Computer Communication Review, vol. 19, No. 4, p. 158-169, David R. Cheriton, Sep. 1989.
  • Protocol Design for High Speed Networks, PhD Thesis, University of Cambridge, Derek Robert McAuley, Sep. 1989.
  • How Slow is One Gigabit Per Second?, ACM Computer Communication Review, vol. 20, No. 1, p. 44-53, Craig Partridge, Jan. 1990.
  • Architectural Considerations for a New Generation of Protocols, ACM Computer Communication Review, vol. 20, No. 4, pp. 200-208, D. D. Clark and D. L. Tennenhouse, Sep. 1990.
  • Protocol Implementation on the Nectar Communication Processor, ACM Computer Communication Review, vol. 20, No. 4, p. 135-144, Eric C. Cooper, et al., Sep. 1990.
  • A Host-Network Interface Architecture for ATM, ACM Computer Communication Review, vol. 21, No. 4, Bruce S. Davie, Sep. 1991.
  • A High-Performance Host Interface for ATM Networks, ACM Computer Communication Review, vol. 21, No. 4, p. 317-325, C. Brendan S. Traw, Sep. 1991.
  • Fairisle: An ATM Network for the Local Area, ACM Computer Communication Review, vol. 21, No. 4, p. 327, Ian Leslie and Derek R. McAuley, Sep. 1991.
  • The Desk Area Network, ACM Operating Systems Review, vol. 25, Issue 4, p. 14-21, Mark Hayter and Derek McAuley, Oct. 1991.
  • An Integration of Network Communication with Workstation Architecture, ACM Computer Communication Review, vol. 21, No. 5, p. 18-29, 1991.
  • The Evolution of XTP, Proceedings of the Third International Conference on High Speed Networking,. Greg Chesson. Nov. 1991.
  • System support for multi-service traffic, University of Cambridge Computer Laboratory Technical Report No. 245, Michael J. Dixon, Jan. 1992.
  • ATOMIC: A Local Communication Network Created Through Repeated Application of Multicomputing Components, Made available by authors, Danny Cohen, Gregory Finn, Robert Felderman, Annette DeSchon, Jan. 10, 1992.
  • Message Authentication with One-Way Hash Functions, ACM Computer Communication Review, vol. 22, No. 5, pp. 29-38, Gene Tsudik, Oct. 1992.
  • Analyzing Communication Latency using the Nectar Communication Processor, ACM Computer Communication Review, vol. 22, No. 4, Peter Steenkiste, Oct. 1992.
  • Efficient Demultiplexing of Incoming TCP Packets, ACM Computer Communication Review, vol. 22, No. 4, Paul E. McKenney and Ken F. Dove, Oct. 1992.
  • TCP/IP on the Parallel Protocol Engine, Proceedings of the IFIP TC6/WG6.4 Fourth International Conference on High Performance Networking IV, Erich Ruetsche and Matthias Kaiserswerth, Dec. 14, 1992.
  • Hardware/Software organization of a high performance ATM host interface, IEEE Journal on Selected Areas in Communications, pp. 240-253, C. Traw and J. Smith, Feb. 1993.
  • The Architecture of Gb/s Multimedia Protocol Adapter, ACM Computer Communication Review, vol. 23, No. 3, E. Ruetsche, Jul. 1993.
  • Giving Applications Access to Gb/s Networking, IEEE Network, vol. 7, Issue 4, pp. 44-52, Jonathan M. Smith and C. Brendan S. Traw, Jul. 1993.
  • The Design and Evaluation of an Off-Host Communications Protocol Architecture, MSci Thesis, University of Virginia, Jeffrey R. Michel, Aug. 1993.
  • A Workstation Architecture to Support Multimedia, PhD Thesis, University of Cambridge, Mark David Hayter, Sep. 1993.
  • The Importance of Non-Data Touching Processing Overheads in TCP/IP, ACM Computer Communication Review, vol. 23, No. 4, pp. 259-268, Jonathan Kay and Joseph Pasquale, Oct. 1993.
  • On the Self-Similar Nature of Ethernet Traffic, ACM Computer Communication Review, vol. 23, No. 4, p. 85-95, W. E. Leland, et al., Oct. 1993.
  • Implementing Network Protocols at User Level, ACM Computer Communication Review, vol. 23, No. 4, C. A. Thekkath, et al., Oct. 1993.
  • A Programmable HIPPI Interface for a Graphics Supercomputer, Proceedings of the 1993 ACM/IEEE conference on Supercomputing, p. 452-461, Portland, Oregon, Raj K. Singh, et al., Nov. 15, 1993.
  • Fbufs: A High-Bandwidth Cross-Domain Transfer Facility, ACM Operating Systems Review, vol. 27, Issue 5, p. 189-202, Peter Druschel and Larry L. Peterson, Dec. 1993.
  • The Parallel Protocol Engine, IEEE/ACM Transactions on Networking, vol. 1, No. 6, p. 650-663, Matthias Kaiserswerth, Dec. 1993.
  • Protocol Service Decomposition for High-Performance Networking, ACM Operating Systems Review, vol. 27, Issue 5, p. 244-255, Chris Maeda, Brian Bershad, Dec. 1993.
  • ETA: Experience with an Intel Xeon Processor as a Packet Processing Engine, IEEE Micro, vol. 24, No. 1, p. 24-31, Greg Regnier, et al., Jan. 1994.
  • A Simple LAN Performance Measure, ACM Computer Communication Review, vol. 24, No. 1, pp. 7-11, J. Vis, Jan. 1994.
  • ATOMIC: A High-Speed Local Communication Architecture, Journal of High Speed Networks, Danny Cohen, Gregory Finn, Robert Felderman, and Annette DeSchon, Jan. 3, 1994.
  • Netstation Architecture Multi-Gigabit Workstation Network Fabric, Proceedings of InterOp '94, Las Vegas, Nevada, Gregory G. Finn and Paul Mockapetris, May 1994.
  • The Medusa Applications Environment, Proceedings of the International Conference on Multimedia Computing and Systems, p. 265-273, Boston, Stuart Wray, et al., May 1994.
  • MPI: A Message-Passing Interface Standard, Message-Passing Interface Forum, University of Tennessee, Knoxville, Various forum members, May 5, 1994.
  • A Programmable Network Interface for a Message-Based Multicomputer, ACM Computer Communication Review, vol. 24, No. 3, p. 8-17, Raj K. Singh, et al., Jul. 1994.
  • Experiences with a High-Speed Network Adaptor: A Software Perspective, ACM Computer Communication Review, vol. 24, No. 4, P. Druschel, et al., Oct. 1994.
  • TCP and Explicit Congestion Notification, ACM, ACM Computer Communication Review, vol. 24, No. 5, p. 8-23, Sally Floyd, Oct. 1994.
  • User-Space Protocols Deliver High Performance to Applications on a Low-Cost Gb/s LAN, ACM Computer Communication Review, vol. 24, No. 4, pp. 14-23, A. Edwards, et al., Oct. 1994.
  • TCP Vegas: New Techniques for Congestion Detection and Avoidance, ACM Computer Communication Review, vol. 24, No. 4, p. 24-35, L. S. Brakmo, et al., Oct. 1994.
  • The Dynamics of TCP Traffic over ATM Networks, ACM Computer Communication Review, vol. 24, No. 4, p. 79-88, A. Romanow and S. Floyd, Oct. 1994.
  • Experiences of Building an ATM Switch for the Local Area, ACM Computer Communication Review, vol. 24, No. 4, p. 158-167, R. J. Black, I. Leslie, and D. McAuley, Oct. 1994.
  • Application-Specific Protocols for User-Level Shared Memory, Proceedings of the 1994 conference on Supercomputing, pp. 380-389, Washington D.C., Babak Falsafi, et al., Nov. 14, 1994.
  • Performance of High-Speed Network I/0 Subsystems: Case Study of A Fibre Channel Network, Proceedings of the 1994 conference on Supercomputing, Washington D.C., Mengjou Lin, et al., Nov. 14, 1994.
  • Myrinet: A Gigabit-per-Second Local-Area Network, Draft of paper published in IEEE Micro, vol. 15, No. 1, pp. 29-36, Nanette J. Boden, et al., Nov. 16, 1994.
  • Beowolf: A Parallel Workstation for Scientific Computation, Proceedings of the 24th International Conference on Parallel Processing, pp. 11-14, Thomas Sterling, et al., Aug. 1995.
  • Software Support for Outboard Buffering and Checksumming, ACM Computer Communication Review, vol. 25, No. 4, p. 87-98, K. Kleinpaste, P. Steenkiste, B. Zill,, Oct. 1995.
  • Performance of Checksums and CRCS over Real Data, ACM Computer Communication Review, vol. 25, No. 4, p. 68-76, C. Partridge, J. Hughes, and J. Stone, Oct. 1995.
  • Experiences Implementing a High-Performance TCP in User-Space, ACM Computer Communication Review, vol. 25, No. 4, A. Edward and S. Muir, Oct. 1995.
  • The Case for Persistent-Connection HTTP, ACM Computer Communication Review, vol. 25, No. 4, J. C. Mogul, Oct. 1995.
  • U-Net: A User-Level Network Interface for Parallel and Distributed Computing, ACM Operating Systems Review, vol. 29, Issue 5, p. 109-126, Thorsten von Eicken, et al., Dec. 1995.
  • Towards an Active Network Architecture, ACM Computer Communication Review, vol. 26, No. 2, pp. 15-20, D. L. Tennenhouse, and D. J. Wetherall, Apr. 1996.
  • Devices in a Multi-Service Operating System, PhD Thesis, University of Cambridge, Paul Ronald Barham, Jul. 1996.
  • Low-Latency Communication on the IBM RISC System/6000 SP, Gigabit Networking Workshop '97 Meeting, Kobe, Japan, Chi-Chao Chang, et al., Nov. 17, 1996.
  • Experiences with a Production Gigabit LAN, Gigabit Networking Workshop '97 Meeting, Kobe, Japan, Joe Touch, et al., Apr. 1997.
  • Host-based Routing Using Peer DMA, Gigabit Networking Workshop '97 Meeting, Kobe, Japan, Joe Touch, et al., Apr. 1997.
  • Report on the 5th IFIP Internation Workshop on Quality of Service (IWQOS 97), ACM Computer Communication Review, vol. 27, No. 3, pp. 100-117, O. Angin, et al., Jul. 1997.
  • Firefly: a Multiprocessor Workstation, ACM Operating Systems Review, vol. 21, Issue 4, p. 164-172, Charles P. Thacker and Lawrence C. Stewart, Oct. 1997.
  • Performance of the CRAY T3E Multiprocessor, Proceedings of the 1997 ACM/IEEE conference on Supercomputing, p. 1-17, San Jose, California, Ed Anderson, et al., Nov. 16, 1997.
  • Performance Evaluation of the SGI Origin2000: A Memory-Centric Characterization of LANL ASCI Applications, Proceedings of the 1997 ACM/IEEE conference on Supercomputing, p. 1-11, San Jose, California, Harvey J. Wassermann, et al., Nov. 16, 1997.
  • An Implementation and Analysis of the Virtual Interface Architecture, Proceedings of the 1998 ACM/IEEE conference on Supercomputing, p. 1-15, Orlando, Florida, Philip Buonadonna, et al., Nov. 7, 1998.
  • MPI-StarT: Delivering Network Performance to Numerical Applications, Proceedings of the 1998 ACM/IEEE conference on Supercomputing, p. 1-15, Orlando, Florida, Parry Husbands and James C. Hoe, Nov. 7, 1998.
  • Avalon: An Alpha/Linux Cluster Achieves 10 Gflops for $150k, Proceedings of the 1998 ACM/IEEE conference on Supercomputing, Orlando, Florida, Michael S. Warren, et al., Nov. 7, 1998.
  • Scaling of Beowulf-class Distributed Systems, Proceedings of the 1998 ACM/IEEE conference on Supercomputing, Orlando, Florida, John Salmon, et al., Nov. 7, 1998.
  • StarT-Voyager: A Flexible Platform for Exploring Scalable SMP Issues, Proceedings of the 1998 ACM/IEEE conference on Supercomputing, Orlando, Florida, Boon S. Ang, et al., Nov. 7, 1998.
  • Enhancing Distributed Systems with Low-Latency Networking, Parallel and Distributed Computing and Networks, Brisbane, Australia, S. L. Pope, et al., Dec. 1998.
  • Internet Vulnerabilities Related to TCP/IP and T/TCP, ACM Computer Communication Review, vol. 29, No. 1, pp. 81-85, M. de Vivo, et al., Jan. 1999.
  • TCP Byte Counting Refinements, ACM Computer Communication Review, vol. 29, No. 3, M. Allman, Jul. 1999.
  • Piglet: A Low-Intrusion Vertical Operating System, Technical Report MS-CIS-00-04, University of Pennsylvania, Steve Muir and Jonathan Smith, Jan. 2000.
  • Characterizing Processor Architectures for Programmable Network Interfaces, Proceedings of the 14th international conference on Supercomputing, pp. 54-65, Santa Fe, New Mexico, Patrick Crowley, et al., May 8, 2000.
  • When The CRC and TCP Checksum Disagree, ACM Computer Communication Review, vol. 30, No. 4, pp. 309-319, Jonathan Stone and Craig Partridge, Oct. 2000.
  • The Failure of TCP in High-Performance Computational Grids, Proceedings of the 2000 ACM/IEEE conference on Supercomputing, Dallas, Texas, W. Feng and P. Tinnakornsrisuphap, Nov. 4, 2000.
  • Architectural and Performance Evaluation of GigaNet and Myrinet Interconnects on Clusters of Small-Scale SMP Servers, Proceedings of the 2000 ACM/IEEE conference on Supercomputing, Dallas, Texas, Jenwei Hsieh, et al., Nov. 4, 2000.
  • Arsenic: A User-Accessible Gigabit Ethernet Interface, Proceedings of IEEE Infocom 2001, pp. 67-76, Ian Pratt and Keir Fraser, Apr. 22, 2001.
  • Deferred Segmentation for Wire-Speed Transmission of Large TCP Frames over Standard GbE Networks, Proceedings of the 9th Symposium on High Performance Interconnects, pp. 81, Bilic Hrvoye, et al., Aug. 22, 2001.
  • Presentation given at HOTT'01, 9th Symposium on High Performance Interconnects, Bilic Hrvoye, et al., Aug. 22, 2001.
  • Topology Discovery For Large Ethernet Networks, ACM Computer Communication Review, vol. 31, No. 4, Bruce Lowekamp, et al., Oct. 2001.
  • EMP: Zero-copy OS-bypass NIC-driven Gigabit Ethernet Message Passing, Proceedings of the 2001 ACM/IEEE conference on Supercomputing, pp. 57, Denver, Piyush Shivam, et al., Nov. 10, 2001.
  • A Case Study in Application I/O on Linux Clusters, Proceedings of the 2001 ACM/IEEE conference on Supercomputing, pp. 11, Denver, Robert Ross, et al., Nov. 10, 2001.
  • On Making TCP More Robust to Packet Reordering, ACM Computer Communication Review, vol. 32, No. 1, E. Blanton and M. Allman, Jan. 2002.
  • TCP Servers: Offloading TCP Processing in Internet Servers. Design, Implementation, and Performance, Technical Report DCR-TR-481, Computer Science Department, Rutgers University, Murali Rangarajan, et al., Mar. 2002.
  • ATM: A Retrospective on Systems Legacy, ACM Computer Communication Review, vol. 32, No. 5, Jon Crowcroft and Derek McAuley, Nov. 2002.
  • A Retrospective View of ATM, ACM Computer Communication Review, vol. 32, No. 5, pp. 13-19, Charles Kalmanek, Nov. 2002.
  • The Influence of ATM on Operating Systems, ACM Computer Communication Review, vol. 32, No. 5, pp. 29-37, Jonathan Smith, Nov. 2002.
  • An Overview of the BlueGene/L Supercomputer, Proceedings of the 2002 ACM/IEEE conference on Supercomputing, pp. 1-22, Baltimore, NR Adiga, et al., Nov. 16, 2002.
  • Ultra-High Performance Communication with MPI and the Sun Fire Link Interconnect, Proceedings of the 2002 ACM/IEEE conference on Supercomputing, p. 1-15, Baltimore, Steven J. Sistare and Christopher J. Jackson, Nov. 16, 2002.
  • Some Internet Architectural Guidelines and Philosophy, IETF Network Working Group, Request for Comments: 3439, R. Bush, D. Meyer, Dec. 2002.
  • F—RTO: An Enhanced Recovery Algorithm for TCP Retransmission Timeouts, ACM Computer Communication Review, vol. 33, No. 2, Pasi Sarolahti, et al., Apr. 2003.
  • Scalable TCP: Improving Performance in Highspeed Wide Area Networks, ACM Computer Communication Review, vol. 33, No. 2, pp. 83-91, Tom Kelly, Apr. 2003.
  • TCP offload is a dumb idea whose time has come, Proceedings of HotOS IX: The 9th Workshop on Hot Topics in Operating Systems, pp. 25-30, Jeffrey C. Mogul, May 18, 2003.
  • A case for Virtual Channel Processors, Proceedings of the ACM SIGCOMM 2003 Workshops, Derek McAuley and Rolf Neugebauer, Aug. 2003.
  • Initial End-to-End Performance Evaluation of 10-Gigabit Ethernet, Proceedings of the 11th Symposium on High Performance Interconnects, Justin Hurwitz and Wu-chun Feng, Aug. 20, 2003.
  • Workshop on network-I/O convergence: experience, lessons, implications (NICELI), ACM Computer Communication Review, vol. 33, No. 5, Vinay Aggarwal, et al., Oct. 2003.
  • Optimizing 10-Gigabit Ethernet for Networks ofWorkstations, Clusters, and Grids: A Case Study, Proceedings of the 2003 ACM/IEEE conference on Supercomputing, Phoenix, Arizona, Wu-chun Feng, et al., Nov. 15, 2003.
  • Performance Comparison of MPI Implementations over InfiniBand, Myrinet and Quadrics, Proceedings of the 2003 ACM/IEEE conference on Supercomputing, Phoenix, Arizona, Jiuxing Liu, et al., Nov. 15, 2003.
  • Architectural Characterization of TCP/IP Packet Processing on the Pentium M Processor, Proceedings of the 10th International Symposium on High Performance Computer Architecture, pp. 152, Srihari Makineni and Ravi Iyer, Feb. 14, 2004.
  • Fast TCP: Motivation, Architecture, Algorithms, Performance, Proceedings of IEEE Infocom, pp. 1246-1259, Cheng Jin, et al., Mar. 7, 2004.
  • TCP Offload to the Rescue, ACM Queue, vol. 2, No. 3, Andy Currid, May 1, 2004.
  • TCP Onloading for Data Center Servers, Computer, IEEE Computer Society, vol. 37, No. 11, pp. 48-58, Greg Regnier, et al., Nov. 2004.
  • Declaration of Dr Gregory L Chesson in Alacritech v. Microsoft, United States District Court, Northern District California, San Francisco Division, Gregory L. Chesson, Feb. 4, 2005.
  • Computing Research: A Looming Crisis, ACM Computer Communication Review, vol. 35, No. 2, Edward D. Lazowska and David A. Patterson, Jul. 2005.
  • Performance Characterization of a 10-Gigabit Ethernet TOE, Proceedings of the 13th Symposium on High Performance Interconnects, W. Feng, et al., Aug. 17, 2005.
  • User-level device drivers: Achieved performance, J. Comput. Sci. & Technol., vol. 20, B. Leslie, et al., Sep. 2005.
  • Head-to-TOE Evaluation of High-Performance Sockets Over Protocol Offload Engines, Proceedings of the IEEE International Conference on Cluster Computing, P. Balaji, et al., Sep. 2005.
  • SCTP versus TCP for MPI, Proceedings of the 2005 ACM/IEEE conference on Supercomputing, Seattle, p. 30, Washington, Humaira Kamal, et al., Nov. 12, 2005.
  • LTCP: Improving the Performance of TCP in Highspeed Networks, ACM Computer Communication Review, vol. 36, No. 1, pp. 41-50, Sumitha Bhandarkar, et al., Jan. 2006.
  • Zero-Copy TCP in Solaris, Proceedings of the USENIX Annual Technical Conference, H. K. Jerry Chu, Jan. 1996.
  • Reflections on Network Architecture: an Active Networking Perspective, ACM Computer Communication Review, vol. 36, No. 2, pp. 27-30, Ken Calvert, Apr. 2006.
  • 10 Networking Papers: Recommended Reading, ACM Computer Communication Review, vol. 36, No. 2, pp. 51-52, Jon Crowcroft, Apr. 2006.
  • Flow labelled IP over ATM: design and rationale, ACM Computer Communication Review, vol. 36, No. 3, pp. 79-92, Greg Minshall, et al., Jul. 2006.
  • 10 Networking Papers: Readings for Protocol Design, ACM Computer Communication Review, vol. 36, No. 3, pp. 77-78, David Wetherall, Jul. 2006.
  • A Critique of RDMA, HPCWire article: http://www.hpcwire.com/features/17886984.html,Patrick Geoffray, Aug. 18, 2006.
  • The NE010 iWARP Adapter, Open Fabrics Alliance, Gary Montry, May 2006.
  • Protocol Onload vs. Offload, Intel, Greg Regnier, Date Unknown.
  • Protocol Onloading vs. Offloading in High Performance Networks, Pacific Northwest Laboratories by Batelle for the U.S. Department of Energy, Fabrizio Petrini, Date Unknown.
  • Thoughts about TCP Offloading, My 2003 HotOS paper title: TCP Offload is a dumd idea whose time has to come, HP Labs (Not speaking for HP), Jeff Mogul, Aug. 2006.
  • Protocol Offloading vs. Onloading in High Performance Networks, Hot Interconnect Panel, System Interconnect Group, Qlogic Corporation, Lloyd Dickman CTO, Aug. 23, 2006.
  • Protocol off-loading vs. on-loading in high performance networks, Hot Interconnects, Myricom®, Patrick Geoffray, 2006.
  • Quantifying the Potential Benefit of Overlapping Communication and Computation in Large-Scale Scientific Applications, Proceedings of the 2006 ACM/IEEE conference on Supercomputing, Tampa, Florida, Jose Carlos Sancho, et al., Nov. 11, 2006.
  • High-Performance and Scalable MPI over InfiniBand with Reduced Memory Usage: An In-Depth Performance Analysis, Proceedings of the 2006 ACM/IEEE conference on Supercomputing, Tampa, Florida, Sayantan Sur, et al., Nov. 11, 2006.
  • 10Gb/s Ethernet Performance and Retrospective, ACM Computer Communication Review, vol. 37, No. 2, pp. 89-92, Steven Pope and David Riddoch, Mar. 19, 2007.
  • Getting 10 Gb/s from Xen, Euro-Par Conference 2007, pp. 224-233, Rennes, France, Kieran Mansley, et al., Aug. 28, 2007.
  • The Parallel Protocol Engine, IEEE/ACM Transactions in Networking vol. 1, Issue 6, pp. 650-663, M. Kaiserswerth, Dec. 1993.
  • Use of message-based multicomputer components to construct gigabit networks, ACM Computer Communication Review, vol. 23, No. 4, p. 32-44, Danny Cohen, et al., Jul. 1993.
  • The End of History, IEEE TCGN Gigabit Networking Workshop, J. Evans and T. Buller, Apr. 22, 2001.
  • The Cambridge Model Distributed System, ACM SIGOPS Operating Systems Review, vol. 14, Issue 1, pp. 21-29, M.V. Wilkes and R.M. Needham, Jan. 1980.
Patent History
Patent number: 8782642
Type: Grant
Filed: Oct 31, 2007
Date of Patent: Jul 15, 2014
Patent Publication Number: 20080065840
Assignee: Solarflare Communications, Inc. (Irvine, CA)
Inventors: Steven Leslie Pope (Costa Mesa, CA), David James Riddoch (Fenstanton), Greg Law (Cambridge)
Primary Examiner: Emerson Puente
Assistant Examiner: Dong Kim
Application Number: 11/982,148
Classifications
Current U.S. Class: Task Management Or Control (718/100); Computer-to-computer Data Routing (709/238); Path Selection (710/38)
International Classification: G06F 9/46 (20060101);